Reconciling meta-learning and continual learning with online mixtures of tasks
This addresses the problem of adapting meta-learning to dynamic or heterogeneous tasks for researchers in few-shot learning, though it is incremental as it builds on existing gradient-based meta-learning and hierarchical Bayes methods.
The paper tackles the challenge of meta-learning when tasks are dissimilar or change over time by proposing a Dirichlet process mixture of hierarchical Bayesian models for task-dependent hyperparameter selection, demonstrating improved handling of latent distribution shift on evolving few-shot learning benchmarks.
Learning-to-learn or meta-learning leverages data-driven inductive bias to increase the efficiency of learning on a novel task. This approach encounters difficulty when transfer is not advantageous, for instance, when tasks are considerably dissimilar or change over time. We use the connection between gradient-based meta-learning and hierarchical Bayes to propose a Dirichlet process mixture of hierarchical Bayesian models over the parameters of an arbitrary parametric model such as a neural network. In contrast to consolidating inductive biases into a single set of hyperparameters, our approach of task-dependent hyperparameter selection better handles latent distribution shift, as demonstrated on a set of evolving, image-based, few-shot learning benchmarks.